Mobile surveys and machine learning can improve urban noise mapping: Beyond A-weighted measurements of exposure.

Urban noise pollution is a major environmental issue, second only to fine particulate matter in its impacts on physical and mental health. To identify who is affected and where to prioritise actions, noise maps derived from traffic flows and propagation algorithms are widely used. These may not reflect true levels of exposure because they fail to consider noise from all sources and may leave gaps where roads or traffic data are absent. We present an improved approach to overcome these limitations. Using walking surveys, we recorded 52,366 audio clips of 10 s each along 733 km of routes throughout the port city of Southampton. We extracted power levels in low (11 to 177 Hz), mid (177 Hz to 5.68 kHz), high (5.68 to 22.72 kHz) and A-weighted frequencies and then built machine-learning (ML) models to predict noise levels at 30 m resolution across the entire city, driven by urban form. Model performance (r ² ) ranged from 0.41 (low frequencies) to 0.61 (mid frequencies) with mean absolute errors of 4.05 to 4.75 dB. The main predictors of noise were related to modes of transport (road, air, rail and water) but for low frequencies, port activities were also important. When mapped to the city scale, A-weighted frequencies produced a similar spatial pattern to mid-frequencies, but did not capture the major sources of low frequency noise from the port or scattered hotspots of high frequencies. We question whether A-weighted noise mapping is adequate for health and wellbeing impact assessments. We conclude that mobile surveys combined with ML offer an alternative way to map noise from all sources and at fine resolution across entire cities that may more accurately reflect true exposures. Our approach is suitable for noise data gathered by citizen scientists, or from a network of sensors, as well as from structured surveys.
Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
(Copyright © 2021 Elsevier B.V. All rights reserved.)